 |
These days there are a lot of research papers coming out on daily basis from the field of computer science which includes solving complex tasks and automating daily workflows based on complex algorithms. Deep learning algorithms in machine learning can do unimaginable things now a days, one of them is Generative Art.
Ai Models
Generative image synthesis models are generating pure synthetic images against the prompt, means when you give some context as text, the AI will generate images according to. There are many algorithms which trained in generative art, most competing 3 models are,
- Dall-E2 by open AI
- Disco Diffusion
- Mid-journey
So i decided to test these 3 for Indian contextual image generation, my concern was to know how well these models understands Indian context and its image synthesis. DallE2 was out on April 6th can only be tested by invite by staying in a wait list, because the model is not published yet, same for Midjourney you have wait to test. Disco-diffusion is open source model and can be used by anyone.
So i enrolled for both Dalle2 (not invited me yet, till this is publishing) and midjourney. Dalle2 waiting lists are tight and selected by many factors, but i got invited to test midjourney after 1 week of sign up for beta. So the post is intended share results of midjourney..
So i enrolled for both Dalle2 (not invited me yet, till this is publishing) and midjourney. Dalle2 waiting lists are tight and selected by many factors, but i got invited to test midjourney after 1 week of sign up for beta. So the post is intended share results of midjourney..
Midjourney
midjourney hit internet on early march 2022 several week before Dall-E2's lauch.
midjourney hit internet on early march 2022 several week before Dall-E2's lauch.
( its unbelievable that models with same concerns are launching around the same date periods, also googles imagen has released after several weeks after dalle2 which also giving stunning photorealistic images on the complex high string prompt same as Dalle2 ) Midjourney beta works as a discord server, where we'll get invited to newbies chat and against the imagine prompt 4 variations will be delivered in 30 seconds, can be requested for the variants and upscaling of those images.
I tested several simple prompts first prompts like
I tested several simple prompts first prompts like
"the coming of super soul in all sectors"
 |
 |
| its quite amusing the AI is capable of understanding the meaning |
 |
"red storm temple in India, having several gods in sky, with thunderstorm and mist at the same time, lots of people worshiping in ground, 4k post-processing"
The temple rendered above seems to be the creation of some northern Indian temple architecture style, looks like ajanta and kailash temples which is always known for its rock-cut-architecture.
 |
Also the AI is picked up rock carving style for creation, the beta of midjourney has around 15~20 test limitation so i never asked it for a variant as it fills the quota.
 |
above prompts are least stringed and i really want to test its limits, so i went for far more complex and vivid prompt If the temples prompt.
"complex Indian god machines surrounded image_artstaion and photo realistic rendering with vivid colors and brilliant lighting"
the result was amusing. if my temples prompt picked the northern style, my god prompt really went to southern idols, i felt the above god look like idol from tirupati or something. So i decided to go for variations.
 |
The color tone was epic, this time its look like a meditative idol sitting in padmasana, the AI must have trained on a large set of data to pick up these minute details. i instantly fell for the colors it rendered. the above render posture feels like a buddha really :-) so i decided to go for variations.....
this time i got a Kurukshethra feeling, like a war machine is coming mood. subs on the both sides as per my prompt has that impact of Kurukshethra or the epic Indian war of ancient times.
this is another variant of temples prompt.
Indian serials are portrayed many of our culture in different scales and perfections, so the above image is feels like beloved Krishna, especially the arch of the crown. As i said its really like muscled war machine like a cyborg of Srikrishna. This render can be taken as model for artists to develop some 3D portraits or devaint style or art-station style render of Krishna the epic..
 |
 |
really i fall for this its that realistic in many ways, especially the carvings, color tones, backgrounds.. its that epic. in a long sight no one ever think it is a rendered or synthesized.
 |
this is the most catchy one in the list, what if our epics are animated or sci-fi-ed in a movie or series or in some comic books (without losing the essence and value) in this type of visual grades? it would be fantastic.....!
Shiv will be suitable for this render, its that manly in this render..
Shiv will be suitable for this render, its that manly in this render..
 |
"festival in Indian temple"
its a lite prompt, only intention was to get a colorful holy like picture, yes rendered ..... its look like a real Indian temple's abstract art.
Curation of Ai Arts
I have some plan to collect and curate these ai generated art works on the go. The collection will largely be on indian context along with others. All of these images will be available under creative common ccby licence soon.
I have some plan to collect and curate these ai generated art works on the go. The collection will largely be on indian context along with others. All of these images will be available under creative common ccby licence soon.
Tags:
Artificial Intelligence
